Model driven architecture (MDA) is an approach to translating real-world algorithms into computer code. It focuses on representing algorithms in a modeling language, followed by a possibly automatic translation of the models into computer code. Doing justice to the definition of MDA’s innovation requires a short, potted history of computers.
Without instructions, or code, computers are just expensive heaters. One of the main problems of the Computer Age has been how to give computers instructions to translate human desire into outputs.
For about 50 years, the answer has been to code in text-based languages, all of which try to aid programming. The first of these text-based languages was assembly, which is more human-readable than machine code, though it still requires a formidable understanding of how registers, memories and, sometimes, pipelined instruction fetches work.
Fortran, which stands for “formula translator” and dates from the 1950s, was the next step in divorcing programmers from a required knowledge of hardware. It hides the computer architecture of registers and presents a syntax that allows humans to concentrate on creating algorithms. Fortran incidentally fortifies the notion of free-floating subroutines and functions.
Other third-generation languages have offered varieties of abstraction from computer hardware, imposing rules and providing new ways for organizing algorithms and data.
Type-safe languages have taken a stab at preventing programmers from making certain mistakes. Permissive languages, which allow easy casting of pointers, have enabled programmers to do all kinds of marvelous things — including stringing up their own coding nooses.
Different approaches to managing algorithms, such as structured programming and object-oriented programming, have been developed. Programming languages that facilitate these paradigms have been created, and libraries of code have been published.
A culture of best practices, which includes requirements analysis, peer reviews and modeling, has arisen to help create software that contains fewer errors and is more efficient, reusable and portable.
Unified Modeling Language (UML) has been promoted by Object Management Group Inc. (OMG), a Needham, Mass.-based standards body that’s maintained by a consortium of interested companies. It has become the language for modeling algorithms and has been adopted by the software community at large.
Originally, an algorithm was expressed in UML before it was manually translated into a text-based language, which was then automatically compiled into assembly language and machine code.
But that raised the question of whether the models themselves could be compiled into machine code, thus improving ways to think about algorithms and to produce software. Efforts to make the process simple and universal were the genesis of the paradigm shift that has led to MDA.
The OMG turned its attention from Common Object Request Broker Architecture to MDA with a white paper in 2000, beginning an effort at classification and standardization and in the process producing a new lexicon, including core notions of the platform-independent model (PIM), platform-specific model (PSM) and Meta-Object Facility (MOF).
UML has a role in MDA, but models do not have to be created in UML to conform to the OMG’s MDA. Instead, both text-based and graphical languages must conform to MOF, which could be called a kind of mother language. (However, it’s hard to say what came first; MOF was abstracted from UML, and UML is one of MOF’s best examples.) MOF is general and universal enough that the disparate companies that make up the OMG do not have to agree to use the same modeling languages. They only have to agree to follow the same principles.
After requirements analysis, modeling in MDA begins with PIMs. These models are meant to capture an algorithm but to ignore the computer hardware and software that will be used to implement them. The models are intended to be pure expressions of the algorithm, with the best division between design and implementation possible.
PIMs are translated into PSMs in the next stage, “compilation,” in which the algorithms are made more ready to run. PSMs have features that correspond to the capabilities of hardware and software. In addition, PIMs may have been “marked” to facilitate this transformation. Finally, PSMs are translated into actual code, which will run on actual hardware, or on actual middleware on actual hardware.
The vision behind MDA is to automate as much of the process of code generation as possible, leaving designers to focus on the algorithm instead. Using tools and languages that will make the design as good as possible, the implementation will come later and will have as little of the pesky interference of fallible humans as possible.
— Matlis is a freelance writer in Newton, Mass. You can reach him at firstname.lastname@example.org.